The Barnum effect

In 1948 Bertram Forer gave a personality test to 39 students and handed each of them a personalised profile. He asked them to rate how well it described them, from zero to five. The average was 4.26.

Every student had received the same profile. Forer had assembled it from a newsstand astrology book.

That is the Barnum effect, named later by Paul Meehl after the showman: a description built out of statements that are true of nearly everyone reads as a precise account of one person. It is the single most useful thing to know about personality content, including ours, because it explains the most common evidence people offer for a result being right — it described me exactly — and shows why that evidence is worth almost nothing.

The conclusion is not that all results are worthless. It is narrower and sharper: the feeling of recognition carries no information about accuracy. Whatever reason you have for trusting a result, it has to be something other than how much it sounded like you.

Forer's paragraph is still in circulation and still works. Its sentences share a small set of properties, and once you can name them you cannot stop seeing them.

Two-sided statements. You have a great deal of unused capacity which you have not turned to your advantage. Both halves flatter, and the claim covers anyone who has ever considered doing more than they are doing.

Paired opposites. At times you are extraverted, affable and sociable, while at other times you are introverted, wary and reserved. This cannot be false. Everyone can find an instance of each, and the reader supplies the ratio.

Socially desirable framing. Snyder and colleagues showed acceptance rises sharply when the content is favourable. A profile that says you are disciplined gets agreement; the same structure saying you are rigid does not.

Universal private experience described as if rare. You have found it unwise to be too frank in revealing yourself to others. Presented as an insight, it is a description of adult life.

A hedge on every specific. Some of your aspirations tend to be pretty unrealistic. Some. Tend. Pretty.

Dickson and Kelly's review found the effect strengthens further when the reader believes the analysis was made for them, when it comes from an authority, and when they participated by answering questions first — which is precisely the sequence a personality test on a website puts you through.

Three ordinary mechanisms do all the work, and none of them involves being credulous. Forer's students were psychology students.

You do the matching. A vague statement is a prompt. Told you find it hard to switch off after a difficult conversation, you retrieve a specific one from last Tuesday. The specificity you experienced was yours, not the profile's. This is why a reading feels more personal the longer you sit with it — you are adding the evidence.

Confirmation is cheap and disconfirmation is expensive. Finding one occasion that fits takes a second. Establishing that a trait statement is false about you requires surveying your own behaviour, which nobody does while reading a web page.

A good feeling gets read as a true one. Favourable content is accepted more readily, and it is not only vanity: a description that flatters produces less resistance, and less resistance feels like fit.

Meehl's point in naming the effect was aimed at clinicians, not horoscopes. He was arguing that a professional report full of Barnum statements looks like clinical insight to the person receiving it and contains no information a statistical formula could not beat. The complaint was about false confidence in interpretation, and it has aged extremely well.

Here is the test, and it takes five seconds.

Could this sentence have come out the other way?

Not is it true of me — a Barnum sentence is true of you, that is the trick. Ask whether the system was capable of telling you the opposite. If the profile for every type would contain some version of the same line, the line is decoration.

Run it on a few examples. You value close relationships but need your own space — no type could have failed to produce that; discard it. You scored at the 81st percentile for conscientiousness, higher than most of this sample — that could have been the 20th, so it carries information, however roughly, and the percentiles and norms entry explains what kind. Your fourth function is inferior Se, which erupts under stress — this sounds falsifiable and is not, for reasons the cognitive function stacks entry sets out.

The same question is the one distinguishing test formats. A result that reports numbers, with a comparison group, on scales that other people score differently on, is making claims that could have been wrong. A result that hands you a paragraph is usually not.

We sell personality tests, so the honest version of this page has to say which of our own outputs would not survive the question above.

Some would. The Big Five reports five scores against a comparison group; any one of them could have come out at the other end, and they frequently do. The screening-style banks report bands that a lot of people do not land in.

Some would not, and we know which. The Colour Personality test, Spirit Animal and Main Character Energy are entertainment. Their outputs are written to be enjoyed, the descriptions are broadly flattering by design, and a reader who finds one of them uncannily accurate has demonstrated the Barnum effect rather than the validity of the instrument. We keep them because they are fun and because they bring people in. That is a commercial reason, not a scientific one, and it does not become a scientific one by being printed next to tests that have a manual behind them.

The type descriptions on our type pages sit in between. The four dimensions the MBTI-style test reports as percentages could each have come out the other way, so those numbers carry information. The prose written over them contains Barnum sentences, as the prose on every type page on the internet does, because that is what makes a type description readable — which means the percentages and the paragraph on the same screen are not equally trustworthy.

None of that is an argument for distrusting everything. It is an argument for reading the numbers and skimming the adjectives — and for noticing that the part of a result which moved you is usually the part that could not have said anything else.